forked from pcyin/NL2code
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathdataset.py
More file actions
1334 lines (979 loc) · 41.6 KB
/
dataset.py
File metadata and controls
1334 lines (979 loc) · 41.6 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
from __future__ import division
import copy
import nltk
from collections import OrderedDict, defaultdict
from query_grammar import QueryParser
import logging
import collections
import numpy as np
import string
import re
import astor
import math
import re
# from canon_utils import fetch_queries_codes, parsable_code, reproducable_code
# from canonicalization import canonicalize_bunch
# from type_simulation import TYPES
# import multiprocessing
# import multiprocessing as mp
from itertools import chain
from nn.utils.io_utils import serialize_to_file, deserialize_from_file
from file_utils import file_contents
import config
from lang.py.parse import get_grammar
from lang.py.unaryclosure import get_top_unary_closures, apply_unary_closures
# define dataset files
ALLANNO = "./en-django/all.anno"
ALLCODE = "./en-django/all.code"
TRAINANNO = "./en-django/train.anno"
DEVANNO = "./en-django/dev.anno"
TESTANNO = "./en-django/test.anno"
#GENALLANNO = "./gendata/pycin_all.anno"
#GENALLCODE = "./gendata/pycin_all.code"
GENALLANNO = "./gendata/all.anno"
GENALLCODE = "./gendata/all.code"
# define actions
APPLY_RULE = 0
GEN_TOKEN = 1
COPY_TOKEN = 2
GEN_COPY_TOKEN = 3
ACTION_NAMES = {APPLY_RULE: 'APPLY_RULE',
GEN_TOKEN: 'GEN_TOKEN',
COPY_TOKEN: 'COPY_TOKEN',
GEN_COPY_TOKEN: 'GEN_COPY_TOKEN'}
class Action(object):
def __init__(self, act_type, data):
self.act_type = act_type
self.data = data
def __repr__(self):
data_str = self.data if not isinstance(self.data, dict) else \
', '.join(['%s: %s' % (k, v) for k, v in self.data.iteritems()])
repr_str = 'Action{%s}[%s]' % (ACTION_NAMES[self.act_type], data_str)
return repr_str
class Vocab(object):
def __init__(self):
self.token_id_map = OrderedDict()
self.insert_token('<pad>')
self.insert_token('<unk>')
self.insert_token('<eos>')
@property
def unk(self):
return self.token_id_map['<unk>']
@property
def eos(self):
return self.token_id_map['<eos>']
def __getitem__(self, item):
if item in self.token_id_map:
return self.token_id_map[item]
logging.debug('encounter one unknown word [%s]' % item)
return self.token_id_map['<unk>']
def __contains__(self, item):
return item in self.token_id_map
@property
def size(self):
return len(self.token_id_map)
def __setitem__(self, key, value):
self.token_id_map[key] = value
def __len__(self):
return len(self.token_id_map)
def __iter__(self):
return self.token_id_map.iterkeys()
def iteritems(self):
return self.token_id_map.iteritems()
def complete(self):
self.id_token_map = dict((v, k) for (k, v) in self.token_id_map.iteritems())
def get_token(self, token_id):
return self.id_token_map[token_id]
def insert_token(self, token):
if token in self.token_id_map:
return self[token]
else:
idx = len(self)
self[token] = idx
return idx
replace_punctuation = string.maketrans(string.punctuation, ' '*len(string.punctuation))
def tokenize(str):
str = str.translate(replace_punctuation)
return nltk.word_tokenize(str)
def gen_vocab(tokens, vocab_size=3000, freq_cutoff=5):
word_freq = defaultdict(int)
for token in tokens:
word_freq[token] += 1
print 'total num. of tokens: %d' % len(word_freq)
words_freq_cutoff = [w for w in word_freq if word_freq[w] >= freq_cutoff]
print 'num. of words appear at least %d: %d' % (freq_cutoff, len(words_freq_cutoff))
ranked_words = sorted(words_freq_cutoff, key=word_freq.get, reverse=True)[:vocab_size-2]
ranked_words = set(ranked_words)
vocab = Vocab()
for token in tokens:
if token in ranked_words:
vocab.insert_token(token)
vocab.complete()
return vocab
class DataEntry:
def __init__(self, raw_id, query, parse_tree, code, actions, meta_data=None):
self.raw_id = raw_id
self.eid = -1
# FIXME: rename to query_token
self.query = query
self.parse_tree = parse_tree
self.actions = actions
self.code = code
self.meta_data = meta_data
@property
def data(self):
if not hasattr(self, '_data'):
assert self.dataset is not None, 'No associated dataset for the example'
self._data = self.dataset.get_prob_func_inputs([self.eid])
return self._data
def copy(self):
e = DataEntry(self.raw_id, self.query, self.parse_tree, self.code, self.actions, self.meta_data)
return e
def __repr__(self):
s = "-" * 60
s = s + "\n" * 2
s = s + str(self.raw_id)
s = s + "\n" * 2
s = s + str(self.eid)
s = s + "\n" * 2
s = s + str(self.query)
s = s + "\n" * 2
s = s + str(self.code)
s = s + "\n" * 2
s = s + str(self.meta_data)
s = s + "\n" * 2
return s
class DataSet:
def __init__(self, annot_vocab, terminal_vocab, grammar, name='train_data', phrase_vocab = None, pos_vocab = None):
self.annot_vocab = annot_vocab
self.terminal_vocab = terminal_vocab
self.name = name
self.examples = []
self.data_matrix = dict()
self.grammar = grammar
self.phrase_vocab = phrase_vocab
self.pos_vocab = pos_vocab
def __repr__(self):
s = "#" * 60
s = s + "\n" * 2
s = s + str(self.annot_vocab)
s = s + "\n" * 2
s = s + str(self.terminal_vocab)
s = s + "\n" * 2
s = s + str(self.phrase_vocab)
s = s + "\n" * 2
s = s + str(self.pos_vocab)
s = s + "\n" * 2
return s
def add(self, example):
example.eid = len(self.examples)
example.dataset = self
self.examples.append(example)
def get_dataset_by_ids(self, ids, name):
dataset = DataSet(self.annot_vocab, self.terminal_vocab,
self.grammar, name)
for eid in ids:
example_copy = self.examples[eid].copy()
dataset.add(example_copy)
for k, v in self.data_matrix.iteritems():
dataset.data_matrix[k] = v[ids]
return dataset
@property
def count(self):
if self.examples:
return len(self.examples)
return 0
def get_examples(self, ids):
if isinstance(ids, collections.Iterable):
return [self.examples[i] for i in ids]
else:
return self.examples[ids]
def get_prob_func_inputs(self, ids):
order = ['query_tokens', 'tgt_action_seq', 'tgt_action_seq_type',
'tgt_node_seq', 'tgt_par_rule_seq', 'tgt_par_t_seq',
'query_tokens_phrase', 'query_tokens_pos']#, 'query_tokens_cid']
max_src_seq_len = max(len(self.examples[i].query) for i in ids)
max_tgt_seq_len = max(len(self.examples[i].actions) for i in ids)
logging.debug('max. src sequence length: %d', max_src_seq_len)
logging.debug('max. tgt sequence length: %d', max_tgt_seq_len)
data = []
for entry in order:
if entry == 'query_tokens':
data.append(self.data_matrix[entry][ids, :max_src_seq_len])
elif entry == 'query_tokens_phrase':
data.append(self.data_matrix[entry][ids, :max_src_seq_len])
elif entry == 'query_tokens_pos':
data.append(self.data_matrix[entry][ids, :max_src_seq_len])
# elif entry == 'query_tokens_cid':
# data.append(self.data_matrix[entry][ids, :max_src_seq_len])
else:
data.append(self.data_matrix[entry][ids, :max_tgt_seq_len])
return data
def init_data_matrices(self, max_query_length=70, max_example_action_num=100):
logging.info('init data matrices for [%s] dataset', self.name)
annot_vocab = self.annot_vocab
terminal_vocab = self.terminal_vocab
phrase_vocab = self.phrase_vocab
pos_vocab = self.pos_vocab
# figure out unique ids for phrase and pos vocab
phrase_vocab_uid = {}
pos_vocab_uid = {}
# fill phrase_vocab unique id
for idx, pv in enumerate(phrase_vocab):
assert (phrase_vocab_uid.get(pv) is None)
phrase_vocab_uid[pv] = idx
# fill pos vocab unique id
for idx, posv in enumerate(pos_vocab):
assert (pos_vocab_uid.get(posv) is None)
pos_vocab_uid[posv] = idx
# np.max([len(e.query) for e in self.examples])
# np.max([len(e.rules) for e in self.examples])
query_tokens = self.data_matrix['query_tokens'] = np.zeros((self.count, max_query_length), dtype='int32')
query_tokens_phrase = self.data_matrix['query_tokens_phrase'] = np.zeros((self.count, max_query_length), dtype='int32')
query_tokens_pos = self.data_matrix['query_tokens_pos'] = np.zeros((self.count, max_query_length), dtype='int32')
query_tokens_cid = self.data_matrix['query_tokens_cid'] = np.zeros((self.count, max_query_length), dtype='int32')
tgt_node_seq = self.data_matrix['tgt_node_seq'] = np.zeros((self.count, max_example_action_num), dtype='int32')
tgt_par_rule_seq = self.data_matrix['tgt_par_rule_seq'] = np.zeros((self.count, max_example_action_num), dtype='int32')
tgt_par_t_seq = self.data_matrix['tgt_par_t_seq'] = np.zeros((self.count, max_example_action_num), dtype='int32')
tgt_action_seq = self.data_matrix['tgt_action_seq'] = np.zeros((self.count, max_example_action_num, 3), dtype='int32')
tgt_action_seq_type = self.data_matrix['tgt_action_seq_type'] = np.zeros((self.count, max_example_action_num, 3), dtype='int32')
for eid, example in enumerate(self.examples):
exg_query_tokens = example.query[:max_query_length]
exg_query_tokens_phrase = example.meta_data['phrase'][:max_query_length]
exg_query_tokens_pos = example.meta_data['pos'][:max_query_length]
exg_action_seq = example.actions[:max_example_action_num]
for tid, token in enumerate(normalize_query_tokens(exg_query_tokens)):
token_id = annot_vocab[token]
query_tokens[eid, tid] = token_id
for tid, token in enumerate(exg_query_tokens):
# if the token contains some _[0-9], thats the canon id
res = re.findall(r"_([0-9]+)", token)
# id
cid = 0
if len(res) == 1:
cid = int(res[0]) + 1
#print token, " 's cid ", cid
# we got the cid
query_tokens_cid[eid, tid] = cid
for tid, p in enumerate(exg_query_tokens_phrase):
assert (phrase_vocab_uid.get(p) is not None)
phrase_id = phrase_vocab_uid[p]
query_tokens_phrase[eid, tid] = phrase_id
for tid, pos in enumerate(exg_query_tokens_pos):
assert (pos_vocab_uid.get(pos) is not None)
pos_id = pos_vocab_uid[pos]
query_tokens_pos[eid, tid] = pos_id
assert len(exg_action_seq) > 0
for t, action in enumerate(exg_action_seq):
if action.act_type == APPLY_RULE:
rule = action.data['rule']
tgt_action_seq[eid, t, 0] = self.grammar.rule_to_id[rule]
tgt_action_seq_type[eid, t, 0] = 1
elif action.act_type == GEN_TOKEN:
token = action.data['literal']
token_id = terminal_vocab[token]
tgt_action_seq[eid, t, 1] = token_id
tgt_action_seq_type[eid, t, 1] = 1
elif action.act_type == COPY_TOKEN:
src_token_idx = action.data['source_idx']
tgt_action_seq[eid, t, 2] = src_token_idx
tgt_action_seq_type[eid, t, 2] = 1
elif action.act_type == GEN_COPY_TOKEN:
token = action.data['literal']
token_id = terminal_vocab[token]
tgt_action_seq[eid, t, 1] = token_id
tgt_action_seq_type[eid, t, 1] = 1
src_token_idx = action.data['source_idx']
tgt_action_seq[eid, t, 2] = src_token_idx
tgt_action_seq_type[eid, t, 2] = 1
else:
raise RuntimeError('wrong action type!')
# parent information
rule = action.data['rule']
parent_rule = action.data['parent_rule']
tgt_node_seq[eid, t] = self.grammar.get_node_type_id(rule.parent)
if parent_rule:
tgt_par_rule_seq[eid, t] = self.grammar.rule_to_id[parent_rule]
else:
assert t == 0
tgt_par_rule_seq[eid, t] = -1
# parent hidden states
parent_t = action.data['parent_t']
tgt_par_t_seq[eid, t] = parent_t
example.dataset = self
class DataHelper(object):
@staticmethod
def canonicalize_query(query):
return query
def parse_django_dataset_nt_only():
from parse import parse_django
#annot_file = '/Users/yinpengcheng/Research/SemanticParsing/CodeGeneration/en-django/all.anno'
annot_file = ALLANNO
vocab = gen_vocab(annot_file, vocab_size=4500)
#code_file = '/Users/yinpengcheng/Research/SemanticParsing/CodeGeneration/en-django/all.code'
code_file = ALLCODE
grammar, all_parse_trees = parse_django(code_file)
train_data = DataSet(vocab, grammar, name='train')
dev_data = DataSet(vocab, grammar, name='dev')
test_data = DataSet(vocab, grammar, name='test')
# train_data
#train_annot_file = '/Users/yinpengcheng/Research/SemanticParsing/CodeGeneration/en-django/train.anno'
train_annot_file = TRAINANNO
train_parse_trees = all_parse_trees[0:16000]
for line, parse_tree in zip(open(train_annot_file), train_parse_trees):
if parse_tree.is_leaf:
continue
line = line.strip()
tokens = tokenize(line)
entry = DataEntry(tokens, parse_tree)
train_data.add(entry)
train_data.init_data_matrices()
# dev_data
#dev_annot_file = '/Users/yinpengcheng/Research/SemanticParsing/CodeGeneration/en-django/dev.anno'
dev_annot_file = DEVANNO
dev_parse_trees = all_parse_trees[16000:17000]
for line, parse_tree in zip(open(dev_annot_file), dev_parse_trees):
if parse_tree.is_leaf:
continue
line = line.strip()
tokens = tokenize(line)
entry = DataEntry(tokens, parse_tree)
dev_data.add(entry)
dev_data.init_data_matrices()
# test_data
#test_annot_file = '/Users/yinpengcheng/Research/SemanticParsing/CodeGeneration/en-django/test.anno'
test_annot_file = TESTANNO
test_parse_trees = all_parse_trees[17000:18805]
for line, parse_tree in zip(open(test_annot_file), test_parse_trees):
if parse_tree.is_leaf:
continue
line = line.strip()
tokens = tokenize(line)
entry = DataEntry(tokens, parse_tree)
test_data.add(entry)
test_data.init_data_matrices()
serialize_to_file((train_data, dev_data, test_data), 'django.typed_rule.bin')
def gen_phrase_vocab(data):
all_phrase = []
for entry in data:
phrase = entry['phrase']
all_phrase = all_phrase + phrase
return list(set(all_phrase))
def gen_pos_vocab(data):
all_pos = []
for entry in data:
pos = entry['pos']
all_pos = all_pos + pos
return list(set(all_pos))
def normalize_query_tokens(tokens):
# consider all TYPES tokens in base form. example STR_0 -> STR
normalized_tokens = []
for t in tokens:
for k, v in TYPES.iteritems():
t = re.sub(v + "_[0-9]+", v, t)
normalized_tokens.append(t)
assert (len(tokens) == len(normalized_tokens))
return normalized_tokens
def parse_django_dataset():
from lang.py.parse import parse_raw
from lang.util import escape
MAX_QUERY_LENGTH = 70
UNARY_CUTOFF_FREQ = 30
##annot_file = '/Users/yinpengcheng/Research/SemanticParsing/CodeGeneration/en-django/all.anno'
##code_file = '/Users/yinpengcheng/Research/SemanticParsing/CodeGeneration/en-django/all.code'
#annot_file = ALLANNO
#code_file = ALLCODE
annot_file = GENALLANNO
code_file = GENALLCODE
#data = preprocess_dataset(annot_file, code_file)
#data = preprocess_gendataset(annot_file, code_file)
data = preprocess_syndataset(annot_file, code_file)
# print data
#for e in data:
# print "-" * 60
# print "\n\n"
# print "idx - ", e['id']
# print "\n\n"
# print "query tokens - ", e['query_tokens']
# print "\n\n"
# print "code - ", e['code']
# print "\n\n"
# print "str_map - ", e['str_map']
# print "\n\n"
# print "raw_code - ", e['raw_code']
# print "\n\n"
# print "bannot - ", e['bannot']
# print "\n\n"
# print "bcode - ", e['bcode']
# print "\n\n"
# print "ref_type - ", e['ref_type']
# print "\n\n"
for e in data:
e['parse_tree'] = parse_raw(e['code'])
parse_trees = [e['parse_tree'] for e in data]
# apply unary closures
# unary_closures = get_top_unary_closures(parse_trees, k=0, freq=UNARY_CUTOFF_FREQ)
# for i, parse_tree in enumerate(parse_trees):
# apply_unary_closures(parse_tree, unary_closures)
# build the grammar
grammar = get_grammar(parse_trees)
# write grammar
with open('django.grammar.unary_closure.txt', 'w') as f:
for rule in grammar:
f.write(rule.__repr__() + '\n')
# # build grammar ...
# from lang.py.py_dataset import extract_grammar
# grammar, all_parse_trees = extract_grammar(code_file)
annot_tokens = list(chain(*[e['query_tokens'] for e in data]))
annot_tokens = normalize_query_tokens(annot_tokens)
annot_vocab = gen_vocab(annot_tokens, vocab_size=5000, freq_cutoff=3) # gen_vocab(annot_tokens, vocab_size=5980)
#annot_vocab = gen_vocab(annot_tokens, vocab_size=5000, freq_cutoff=0) # gen_vocab(annot_tokens, vocab_size=5980)
terminal_token_seq = []
empty_actions_count = 0
# helper function begins
def get_terminal_tokens(_terminal_str):
# _terminal_tokens = filter(None, re.split('([, .?!])', _terminal_str)) # _terminal_str.split('-SP-')
# _terminal_tokens = filter(None, re.split('( )', _terminal_str)) # _terminal_str.split('-SP-')
tmp_terminal_tokens = _terminal_str.split(' ')
_terminal_tokens = []
for token in tmp_terminal_tokens:
if token:
_terminal_tokens.append(token)
_terminal_tokens.append(' ')
return _terminal_tokens[:-1]
# return _terminal_tokens
# helper function ends
# first pass
for entry in data:
idx = entry['id']
query_tokens = entry['query_tokens']
code = entry['code']
parse_tree = entry['parse_tree']
for node in parse_tree.get_leaves():
if grammar.is_value_node(node):
terminal_val = node.value
terminal_str = str(terminal_val)
terminal_tokens = get_terminal_tokens(terminal_str)
for terminal_token in terminal_tokens:
assert len(terminal_token) > 0
terminal_token_seq.append(terminal_token)
terminal_vocab = gen_vocab(terminal_token_seq, vocab_size=5000, freq_cutoff=3)
#terminal_vocab = gen_vocab(terminal_token_seq, vocab_size=5000, freq_cutoff=0)
phrase_vocab = gen_phrase_vocab(data)
pos_vocab = gen_pos_vocab(data)
#assert '_STR:0_' in terminal_vocab
train_data = DataSet(annot_vocab, terminal_vocab, grammar, 'train_data', phrase_vocab, pos_vocab)
dev_data = DataSet(annot_vocab, terminal_vocab, grammar, 'dev_data', phrase_vocab, pos_vocab)
test_data = DataSet(annot_vocab, terminal_vocab, grammar, 'test_data', phrase_vocab, pos_vocab)
all_examples = []
can_fully_gen_num = 0
# second pass
for entry in data:
idx = entry['id']
query_tokens = entry['query_tokens']
code = entry['code']
str_map = entry['str_map']
parse_tree = entry['parse_tree']
rule_list, rule_parents = parse_tree.get_productions(include_value_node=True)
#print "Rule List - "
#for r in rule_list:
# print "Rule -", r
#for k, v in rule_parents.iteritems():
# print "Rule parents - ", k, " - ", v
#print "Rule parents - ", rule_parents
actions = []
can_fully_gen = True
rule_pos_map = dict()
for rule_count, rule in enumerate(rule_list):
if not grammar.is_value_node(rule.parent):
assert rule.value is None
parent_rule = rule_parents[(rule_count, rule)][0]
if parent_rule:
parent_t = rule_pos_map[parent_rule]
else:
parent_t = 0
rule_pos_map[rule] = len(actions)
d = {'rule': rule, 'parent_t': parent_t, 'parent_rule': parent_rule}
action = Action(APPLY_RULE, d)
actions.append(action)
else:
assert rule.is_leaf
parent_rule = rule_parents[(rule_count, rule)][0]
parent_t = rule_pos_map[parent_rule]
terminal_val = rule.value
terminal_str = str(terminal_val)
terminal_tokens = get_terminal_tokens(terminal_str)
# assert len(terminal_tokens) > 0
for terminal_token in terminal_tokens:
term_tok_id = terminal_vocab[terminal_token]
tok_src_idx = -1
try:
tok_src_idx = query_tokens.index(terminal_token)
except ValueError:
pass
d = {'literal': terminal_token, 'rule': rule, 'parent_rule': parent_rule, 'parent_t': parent_t}
# cannot copy, only generation
# could be unk!
if tok_src_idx < 0 or tok_src_idx >= MAX_QUERY_LENGTH:
action = Action(GEN_TOKEN, d)
if terminal_token not in terminal_vocab:
if terminal_token not in query_tokens:
# print terminal_token
can_fully_gen = False
else: # copy
if term_tok_id != terminal_vocab.unk:
d['source_idx'] = tok_src_idx
action = Action(GEN_COPY_TOKEN, d)
else:
d['source_idx'] = tok_src_idx
action = Action(COPY_TOKEN, d)
actions.append(action)
d = {'literal': '<eos>', 'rule': rule, 'parent_rule': parent_rule, 'parent_t': parent_t}
actions.append(Action(GEN_TOKEN, d))
if len(actions) == 0:
empty_actions_count += 1
continue
example = DataEntry(idx, query_tokens, parse_tree, code, actions,
{'raw_code': entry['raw_code'], 'str_map': entry['str_map'],
'phrase' : entry['phrase'], 'pos' : entry['pos'],
'bannot' : entry['bannot'], 'bcode' : entry['bcode'],
'ref_type' : entry['ref_type']})
if can_fully_gen:
can_fully_gen_num += 1
# train, valid, test
if 0 <= idx < 13000:
train_data.add(example)
elif 13000 <= idx < 14000:
dev_data.add(example)
else:
test_data.add(example)
# modified train valid test counts
#if 0 <= idx < 10000:
# train_data.add(example)
#elif 10000 <= idx < 11000:
# dev_data.add(example)
#else:
# test_data.add(example)
all_examples.append(example)
# print statistics
max_query_len = max(len(e.query) for e in all_examples)
max_actions_len = max(len(e.actions) for e in all_examples)
serialize_to_file([len(e.query) for e in all_examples], 'query.len')
serialize_to_file([len(e.actions) for e in all_examples], 'actions.len')
logging.info('examples that can be fully reconstructed: %d/%d=%f',
can_fully_gen_num, len(all_examples),
can_fully_gen_num / len(all_examples))
logging.info('empty_actions_count: %d', empty_actions_count)
logging.info('max_query_len: %d', max_query_len)
logging.info('max_actions_len: %d', max_actions_len)
train_data.init_data_matrices()
dev_data.init_data_matrices()
test_data.init_data_matrices()
#print train_data
## print train_data matrix
#print "Data matrix: query_tokens "
#print train_data.data_matrix['query_tokens']
#print "\n" * 2
#print "Data matrix : query_tokens_phrase"
#print "\n" * 2
#print train_data.data_matrix['query_tokens_phrase']
#print "\n" * 2
#print "Data matrix : query_tokens_pos"
#print "\n" * 2
#print train_data.data_matrix['query_tokens_pos']
#print "\n" * 2
#print "Data matrix : query_tokens_cid"
#print "\n" * 2
#print train_data.data_matrix['query_tokens_cid']
#print "\n" * 2
## print few data entries
#for d in train_data.examples[:5]:
# print "\n" * 2
# print d
## lets print dataset for good measure
serialize_to_file((train_data, dev_data, test_data),
# 'data/django.pnet.qparse.dataset.freq3.par_info.refact.space_only.bin')
'data/django.pnet.fullcanon.dataset.freq3.par_info.refact.space_only.bin')
# 'data/django.pnet.dataset.freq3.par_info.refact.space_only.bin')
#'data/django.cleaned.dataset.freq3.par_info.refact.space_only.order_by_ulink_len.bin')
# 'data/django.cleaned.dataset.freq5.par_info.refact.space_only.unary_closure.freq{UNARY_CUTOFF_FREQ}.order_by_ulink_len.bin'.format(UNARY_CUTOFF_FREQ=UNARY_CUTOFF_FREQ))
return train_data, dev_data, test_data
def check_terminals():
from parse import parse_django, unescape
#grammar, parse_trees = parse_django('/Users/yinpengcheng/Research/SemanticParsing/CodeGeneration/en-django/all.code')
#annot_file = '/Users/yinpengcheng/Research/SemanticParsing/CodeGeneration/en-django/all.anno'
grammar, parse_trees = parse_django(ALLCODE)
annot_file = ALLANNO
unique_terminals = set()
invalid_terminals = set()
for i, line in enumerate(open(annot_file)):
parse_tree = parse_trees[i]
utterance = line.strip()
leaves = parse_tree.get_leaves()
# tokens = set(nltk.word_tokenize(utterance))
leave_tokens = [l.label for l in leaves if l.label]
not_included = []
for leaf_token in leave_tokens:
leaf_token = str(leaf_token)
leaf_token = unescape(leaf_token)
if leaf_token not in utterance:
not_included.append(leaf_token)
if len(leaf_token) <= 15:
unique_terminals.add(leaf_token)
else:
invalid_terminals.add(leaf_token)
else:
if isinstance(leaf_token, str):
print leaf_token
# if not_included:
# print str(i) + '---' + ', '.join(not_included)
# print 'num of unique leaves: %d' % len(unique_terminals)
# print unique_terminals
#
# print 'num of invalid leaves: %d' % len(invalid_terminals)
# print invalid_terminals
def query_to_data(query, annot_vocab):
query_tokens = query.split(' ')
token_num = min(config.max_query_length, len(query_tokens))
data = np.zeros((1, token_num), dtype='int32')
for tid, token in enumerate(query_tokens[:token_num]):
token_id = annot_vocab[token]
data[0, tid] = token_id
return data
QUOTED_STRING_RE = re.compile(r"(?P<quote>['\"])(?P<string>.*?)(?<!\\)(?P=quote)")
def canonicalize_query(query):
"""
canonicalize the query, replace strings to a special place holder
"""
str_count = 0
str_map = dict()
matches = QUOTED_STRING_RE.findall(query)
# de-duplicate
cur_replaced_strs = set()
for match in matches:
# If one or more groups are present in the pattern,
# it returns a list of groups
quote = match[0]
str_literal = quote + match[1] + quote
if str_literal in cur_replaced_strs:
continue
# FIXME: substitute the ' % s ' with
if str_literal in ['\'%s\'', '\"%s\"']:
continue
str_repr = '_STR:%d_' % str_count
str_map[str_literal] = str_repr
query = query.replace(str_literal, str_repr)
str_count += 1
cur_replaced_strs.add(str_literal)
# tokenize
query_tokens = nltk.word_tokenize(query)
new_query_tokens = []
# break up function calls like foo.bar.func
for token in query_tokens:
new_query_tokens.append(token)
i = token.find('.')
if 0 < i < len(token) - 1:
new_tokens = ['['] + token.replace('.', ' . ').split(' ') + [']']
new_query_tokens.extend(new_tokens)
query = ' '.join(new_query_tokens)
return query, str_map
def canonicalize_example(query, code):
from lang.py.parse import parse_raw, parse_tree_to_python_ast, canonicalize_code as make_it_compilable
import astor, ast
canonical_query, str_map = canonicalize_query(query)
canonical_code = code
for str_literal, str_repr in str_map.iteritems():
canonical_code = canonical_code.replace(str_literal, '\'' + str_repr + '\'')
canonical_code = make_it_compilable(canonical_code)
# sanity check
parse_tree = parse_raw(canonical_code)
gold_ast_tree = ast.parse(canonical_code).body[0]
gold_source = astor.to_source(gold_ast_tree)
ast_tree = parse_tree_to_python_ast(parse_tree)
source = astor.to_source(ast_tree)
assert gold_source == source, 'sanity check fails: gold=[%s], actual=[%s]' % (gold_source, source)
query_tokens = canonical_query.split(' ')
return query_tokens, canonical_code, str_map
def process_query(query, code):
from parse import code_to_ast, ast_to_tree, tree_to_ast, parse
import astor
str_count = 0
str_map = dict()
match_count = 1
match = QUOTED_STRING_RE.search(query)
while match:
str_repr = '_STR:%d_' % str_count
str_literal = match.group(0)
str_string = match.group(2)
match_count += 1
# if match_count > 50:
# return
#
query = QUOTED_STRING_RE.sub(str_repr, query, 1)
str_map[str_literal] = str_repr
str_count += 1
match = QUOTED_STRING_RE.search(query)
code = code.replace(str_literal, '\'' + str_repr + '\'')
# clean the annotation
# query = query.replace('.', ' . ')
for k, v in str_map.iteritems():
if k == '\'%s\'' or k == '\"%s\"':
query = query.replace(v, k)
code = code.replace('\'' + v + '\'', k)
# tokenize
query_tokens = nltk.word_tokenize(query)
new_query_tokens = []
# break up function calls
for token in query_tokens:
new_query_tokens.append(token)
i = token.find('.')
if 0 < i < len(token) - 1:
new_tokens = ['['] + token.replace('.', ' . ').split(' ') + [']']
new_query_tokens.extend(new_tokens)
# check if the code compiles
tree = parse(code)
ast_tree = tree_to_ast(tree)
astor.to_source(ast_tree)
return new_query_tokens, code, str_map
def deoneline(txt):
orig_line = ""
return txt.replace(" DCNL ", '\n').replace(" DCSP ", '\t')
def preprocess_gendataset(annot_file, code_file):
f_annot = open('annot.all.canonicalized.txt', 'w')
f_code = open('code.all.canonicalized.txt', 'w')
# get all queries
contents = file_contents(annot_file)
examples = contents.split("\n\n")
examples = examples[:-1]
queries = []
for ex in examples: